locating and tracking ble beacons with smartphones

Locating and Tracking BLE Beacons with SmartphonesDongyao Chen

∗, Kang G. Shin

University of Michigan, Ann Arbor

{chendy,kgshin}@umich.edu

Yurong Jiang∗, Kyu-Han Kim

Hewlett Packard Labs

{yurong.jiang,kyu-han.kim}@hpe.com

ABSTRACTWe present a smartphone-based application, called LocBLE, for en-

abling users to estimate the location of nearby Bluetooth low energy

(BLE) beacons. In contrast to existing BLE beacon-based proximity

applications that can only show coarse-grained (immediate, near,

and far) distance estimation, LocBLE’s ne-grained estimation can

enhance human-environment interactions.

LocBLE has three salient features in estimating location from

BLE beacon signals. First, it is adaptive to dynamic signal propaga-

tion environments by learning the environmental changes directly

from the received signal strength (RSS). Second, it performs sensor-

fusion for location estimation by utilizing motion sensor data and

RSS readings from a smartphone. Finally, LocBLE improves location

tracking accuracy with novel on-line calibration on a set of bea-

cons nearby. We have built a prototype of LocBLE on smartphones

and evaluated it on commodity proximity-enabled beacons. Our

experimental results demonstrate that LocBLE achieves an average

of 1.8m and 1.2m accuracies in locating indoor and outdoor BLE

beacons, respectively.

CCS CONCEPTS•Human-centered computing→Ubiquitous andmobile com-puting;

KEYWORDSBluetooth low energy, Internet of Things (IoT)

1 INTRODUCTIONBluetooth Low Energy (BLE) beacons have become very popular

with numerous emerging applications of IoT (Internet-of-Thing),

AR (Augmented Reality), and home automation. Proximity estima-

tion, as one of the most representative features, is expected to ex-

pand human–environment interactions in various applications, in-

cluding retail marketing [1], health-care [2], and transportation [3].

For example, BLE beacons have already been deployed in many

retail giants, including Target andMacy’s, showing the proximity of

items on the customers’ phones and thus creating a more engaging

shopping experience [4].

Permission to make digital or hard copies of all or part of this work for personal or

classroom use is granted without fee provided that copies are not made or distributed

for prot or commercial advantage and that copies bear this notice and the full citation

on the rst page. Copyrights for components of this work owned by others than ACM

must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,

to post on servers or to redistribute to lists, requires prior specic permission and/or a

fee. Request permissions from [email protected].

CoNEXT ’17, December 12–15, 2017, Incheon, Republic of Korea© 2017 Association for Computing Machinery.

ACM ISBN 978-1-4503-5422-6/17/12. . . $15.00

https://doi.org/10.1145/3143361.3143385

(a) Finding item

Interesteditem

(b) AR tagging

Figure 1: Featured use-cases of ne-grained location of BLEbeacons.

Even though proximity beacons have been pervasively deployed,

their functionalities are limited to a few “nearable” applications.

Specically, existing applications can only provide a rough proxim-

ity information.1This feature hampers the usability of BLE beacons

in various scenarios, especially those requiring (2-dimensional) po-

sition information of beacons. Let’s highlight two representative

use-cases. Fig. 1(a) shows use of smartphones to nd a lost item by

locating an attached BLE beacon on the lost item, while Fig. 1(b)

shows AR users’ concentration on the items of their interest that

are highlighted by an attached BLE beacon, thus helping them focus

on items even when they can’t see due to the blockage of line of

sight.

The cause of limiting existing BLE beacon’s usability is rooted

at its low-power design. As we will elaborate in Sec. 2, to achieve

the longest possible battery life, the BLE protocol design species a

series of power-saving features, including low transmission power,

narrow bandwidth, simplied connection design, and low duty

cycle. Due to this ultra light-weighted protocol design, the received

signal strength (RSS) of BLE beacon’s advertisement becomes the

only indicator for distance estimation. Unlike other popular location

indicators, such as channel state information (CSI) [5, 6], time-

of-ight, and angle-of-arrival, RSS readings are shown to exhibit

large uctuations due to dynamic propagation environments [6–8].

These features together make it challenging to locate and track BLE

beacons.

To address this challenge, we present LocBLE, a low-cost smartphone-

based application for estimating location of nearby commodity BLE

beacons. In its core, the design of LocBLE is comprised of three

novel components to mitigate RSS uctuations and extract location

information from limited resources.

First, since RSS reading is susceptible to environment changes,

to make LocBLE adaptive to dynamic environments, we need to

answer a challenging question “how to make LocBLE aware of

*The authors equally contributed to this work. This work was initiated during the

rst author’s internship with Hewlett Packard Labs

11-dimensional, four proximity zones (immediate, near, far, and unknown)

https://doi.org/10.1145/3143361.3143385

CoNEXT ’17, December 12–15, 2017, Incheon, Republic of Korea D. Chen et al.

environmental changes?” To answer this question, we design a

novel environmental estimation module, called EnvAware. This

module infers environmental changes from RSS readings via a SVM-

based algorithm. This estimation of environmental changes enables

us to understand the varying channel condition, and allows LocBLE

to adjust the following location estimation when the environment

undergoes signicant changes.

Second, with awareness of environmental changes, LocBLE needs

to estimate the target beacon’s location based on RSS readings. This

task is challenging for two reasons: 1) the traditional log-based path-

loss model of radio frequency (RF) signal is formulated for only

calculating 1-dimensional distance [9]; 2) the parameters in the log-

based model uctuates due to dierent environments and hardware

congurations. We address both challenges by introducing a novel

data fusion scheme based on inertial sensor data and RSS readings

on smartphones. To estimate location, our algorithm uses a reverse

regression and only requires the user to make a short movement

for measurements. To adapt to changing parameters, our algorithm

estimates the parameter set, instead of using constant numbers.

Unlike existing estimation schemes [10, 11], LocBLE employs a

novel path loss exponent estimation based on data fusion of RSS

readings and motion sensors for path loss exponent estimation on

mobile devices.

Finally, to explore opportunities for rening LocBLE’s accuracy,

we propose a novel way to rene location estimation by locating

multiple neighboring BLE beacons together. Our design is inspired

by an observation that multiple BLE beacons are likely close to each

other. For example, in a retail store, items of the same category are

stocked together. We exploit this and propose a novel clustering

algorithm for improving accuracy. Specically, LocBLE uses an

algorithm based on dynamic time warping (DTW), allowing further

calibration of location estimation.

We have built a prototype of LocBLE on both iOS and Android

platforms. We have evaluated LocBLE using commodity proximity

BLE beacons in various (8 indoor and 1 outdoor) environments.

LocBLE is shown to achieve an average of 1.8m indoor and 1.2m

outdoor accuracies in estimating BLE locations.

This paper makes the following three contributions:

• Development of LocBLE, a system for estimating BLE beacon

location. LocBLE runs on smartphones and is fully functional

with commodity BLE beacons.

• Introduction of three key design elements for enabling LocBLE

adaptable to various environments (Sec. 4, Sec. 5, Sec. 6);

• Implementation and evaluation of LocBLE on commodity

smartphones and BLE beacons (Sec. 7).

2 BACKGROUND2.1 Nearable TechnologiesRecently, nearable technology has received signicant attention for

its potential for enriching human–environment interactions. Exist-

ing RFID, NFC tags, and BLE beacons have shown their capability

of practical asset tracking. However, RFID tag tracking [12, 13]

requires a dedicated RFID reader, making it infeasible on light-

weighted devices (e.g. smartphones). NFC tags can be detected by

smartphones, but their range is limited to tens of centimeters [14].

In contrast to these technologies, BLE beacons are compatible with

commodity mobile devices, and have wider coverage (8–15m in

indoor settings).

2.2 BLE Beacon PrimerThe BLE beacon represents a class of BLE devices, and advertises its

identier to nearby devices by broadcasting a BLE signal. To enable

both portability and long battery life, BLE beacons are usually

powered by coin cell batteries and operate with a power-saving

design to guarantee 1–3 years of lifetime. However, this ultra low-

power design limits the resource that can be used for estimating

the location of a BLE beacon.

Limited transmission power. BLE limits the transmission power

to reduce energy consumption. BLE v4.0, v4.1, and v4.2 dened

the maximum output power to be 10mW, which is 10x lower than

WiFi transmission power specied by the FCC [15]. This feature

constrains the BLE beacon’s range ( <15m indoor) and makes the

BLE beacon signal more susceptible to path loss caused by the

blockage of signal propagation [9]. Note that the upcoming BLE

v5.0 sets the maximum output power to 100mW, but this high Tx

power is designed exclusively for high power devices with Class 1

BLE chip [16].

Narrow bandwidth and frequency hopping. To coexist with

WiFi and other RF signals that operate in the 2.4GHz frequency

band, BLE specications incorporate narrow bandwidth and fre-

quency hopping. Specically, in advertisement state, a BLE device

hops in a xed sequence of 3 dedicated channels (37, 38, and 39, each

with 2MHz bandwidth). In connection state, the frequency changes

pseudorandomly among 40 channels [17]. These features make the

BLE signal more susceptible to frequency-selective fading [6].

Connectivity and advertisement of BLEbeacons.According toBLE beacon’s specication of connectivity [17], a connectable BLE

beaconworks as a BLE peripheral device (e.g., Bluetoothmouse) and

can receive pairing requests, whereas a non-connectable beacon is

essentially a BLE device that works only in broadcasting mode. To

determine the connectivity of a targeted BLE beacon, the receiving

device can inspect the connectivity type indicated by the rst 4 bits

in the header advertising channel protocol data units (PDUs). Inter-

ested readers are referred to the BLE specication (page 2567, [17])

for more details. The non-connectible mode of BLE beacons can ex-

tend battery life by limiting the interaction between the peripheral

(e.g., BLE beacons) and central (e.g., smartphones that are scanning

their surrounding beacons) devices. To reduce power consumption

further, the duty cycles for broadcasting advertisement are limited

to be under 100ms and 20ms on non-connectable and connectable

beacons, respectively [18].

To be compatible with the low power design of existing com-

modity beacons, LocBLE focuses on locating non-connectable BLE

beacons.

Locating and Tracking BLE Beacons with Smartphones CoNEXT ’17, December 12–15, 2017, Incheon, Republic of Korea

0 1.5 3.0 4.6 6.1−100

−90

−80

−70

−60

−50

−40

Distance (m)

RSSI

(dBm

)

iPhone 5sNexus 5xMoto Nexus 6iPhone 5sNexus 5xMoto Nexus 6

Figure 2: RSS reading on dierent smartphones.

2.3 RF Signal FadingLike other RF signals, the BLE beacon signal suers from signal

degradation. Multipath fading occurs when RF signals reach the

receiving antenna via multiple dierent paths. The dierent lengths

of these paths make the received signals combined constructively

or destructively. This eect further exacerbates the BLE signal’s

strength.

To cope with signal fading, existing WiFi and cellular networks

use CSI feedback to adjust Tx power. Even though the manufac-

turer can replace the subset of an advertisement packet with a

training symbol to enable CSI feedback, existing BLE beacons, such

as iBeacon, EddyStone, and AltBeacon, are not compatible with

this feature. For compatibility with existing BLE beacons without

modication, we propose a new environment-aware method (Sec. 4)

for adjusting the estimation of BLE beacon location.

2.4 RSS Measurement at ReceiverRSS measurements are also aected by the receiver’s hardware

conguration. Specically, noises will be added to RSS readings

due to the CMOS property of analog components, imperfections,

and environment temperature. For example, the widely-used Broad-

Com BCM4334 WLAN/Bluetooth receiver chipset [19] has ±5 RSS

accuracy at room temperature. This oset is another source of the

noise in RSS readings. To mitigate this impact, we propose a novel

energy-oset estimation scheme (Sec. 5).

2.5 RSS for Location EstimationTo evaluate the reliability of using RSS for BLE beacon location

estimation, we rst collect RSS readings on dierent smartphones in

real-life indoor environments. In our experiment, we use 3 dierent

smartphones, and walk away from the target BLE beacon on the

same path. As shown in Fig. 2, despite the changes of data osets

on dierent smartphones, the RSS trend shows the same pattern. To

mitigate the impact of RSS uctuations further, we extract location

information from the changing trend of RSS readings. Next, we will

detail the design of LocBLE.

3 OVERVIEWFig. 3 shows the three-layer system architecture of LocBLE: datacollection, location estimation, and calibration.

Data collection layer. This layer collects and processes various

sensory data, including magnetometer, IMU, and RSS readings from

BLE API (e.g., CoreBluetooth in iOS, getBluetoothLeScanner in

Android).

Location estimation layer. This layer estimates the relative lo-

cation of target BLE beacons. It has three main functionalities: 1)

mitigating RSS uctuations with both environment recognition (En-

vAware) and adaptive noise ltering (ANF) (Sec. 4); 2) measuring

the observer’s movement by processing the motion sensor and mag-

netometer readings (Sec. 5); and 3) estimating the target’s location

by fusing measured RSS and motion data (Sec. 5).

Calibration layer. This layer explores opportunities for improving

estimation accuracy. If there are multiple beacons with similar

location estimation (or located nearby), their corresponding RSS

readings often show a similar changing pattern. Based on this

observation, the calibration layer rst recognizes whether there are

multiple beacons nearby, and then rene the location estimation

with a probabilistic weight algorithm (Sec. 6).

Data collection layer Location estimation layer Calibration layer

Nodesclustering

Adaptivecalibration

Is targetmoving?

Noisefiltering

Datafusio

n

EnvAware

Motiontracker

MotionSensor

BLE Scanner

Magnetometer

Figure 3: System architecture of LocBLE.

4 DATA PREPROCESSING BASED ONENVIRONMENTAL CHANGES

The key reason for distorted RSS readings is the channel uctuations

caused by environmental changes. So, it is essential to preprocess

RSS data for further analysis as we will show in Secs. 5 and 6. Specif-

ically, LocBLE takes a two-step approach to preprocess RSS data: 1)

recognizes environmental changes directly from RSS readings, and

2) adaptive noise ltering to smooth RSS data.

4.1 Environment recognitionDespite uctuating RSS values, LocBLE uses the changing trend of

RSS to estimate the target location. For a given model, our estima-

tion will become the more accurate with more data. However, RSS

readings may also be aected by environmental changes during

the data collection process, thus yielding less accurate results. To

address this problem, we propose an adaptive estimation method

called EnvAware: it recognizes current environmental changes and

use it to tune location estimation as discussed in Sec. 5.

Feature extraction and classiers. Our RSS feature extraction

segments the signal values into short (1–2s) windows and operates

on them. Specically, our feature vector comprised by the statistics

of a new time window vector V : mean, variance, skewness. Beside


these statistics, we also use 5 values directly from V : minimum,

rst quartile, median, third quartile, and max value. Finally, our

feature vector is composed of the standardized 9 values described

above. Such a feature denition turns out to be the most accurate for

the various classiers we tried: SVM with various kernels, DecisionTree Classier, RandomForest Classier, etc. In LocBLE, we chose

SVM with a linear kernel as our classier since it outperforms other

algorithms in the ensemble.

To construct an RSS dataset by accounting for real-world signal

propagation, we collected RSS data on smartphones in three rep-

resentative environments: line-of-sight (LOS), partial-line-of-sight

(p-LOS), non-line-of-sight (NLOS). p-LOS represents the propaga-

tion scenario that has blockage with a low blocking coecient,

such as glass, wooden door, and human body, etc., while NLOS

represents the propagation scenario that has blockage with a high

blocking coecient, such as concrete wall, cinder wall, and metal

board, etc. LocBLE aims to classify these three types of environment.

Each data trace in our dataset was labeled with the corresponding

environment. Our SVM-based algorithm is implemented by using

sklearn module [20] in Python.

We collected training data in each designed type and labeled

them according to the above 3 environment categories. For instance,

for the blocked type, we placed one device behind a blocking object,

the other device stores all the RSS data while moving around in

front of the object. We also varied the blocking object, like wall,

human body, etc. We used a time window of 2 seconds to generate

the feature vector. Our classication achieved an excellent classi-

cation accuracy (94.7% precision and 94.5% recall for our three-type

classication).

To incorporate EnvAware with our location estimation scheme

(Sec. 5), LocBLE keeps monitoring environmental changes, and

starts a new regression model only if new incoming data shows

abrupt environmental changes.

4.2 Adaptive Noise FilteringTo smooth RSS data for next-step processing, LocBLE passes raw

RSS data through an adaptive noise lter (ANF), and ANF is based

on two noise ltering techniques: 1) a ne-tuned Butterworth lter,

and 2) adaptive Kalman lter (AKF).

Butterworth lter design. To remove the eect of fast fading

caused by environmental changes and device movements, we de-

signed a low-pass lter based on a 6th-order Butterworth lter

(BF).

Design of AKF. Our BF design can smooth uctuating RSS data.

However, the high order of BF also introduces delay and under-

mines the responsiveness of ltered data. To mitigate the impact of

delay, we propose AKF, a modied Kalman lter. AKF enhances the

responsiveness of lter by fusing raw RSS readings with BF output.

More details can be found in [21].

Fig. 4 illustrates an example of BF + AKF processing results. BF

achieves a much smoother result by ltering raw data, but it adds

delay and is not fast in responding to RSS changes. We then apply

AKF to achieve better performance than using BF alone.

0 10 20 30 40−90

−85

−80

−75

−70

−65

Time stamp (s)

RSS

I (dB

m)

TheoreticalBPF + AKFRawBPF

32 34 36 38−77

−76

−75

−74

Time stamp

RSSI

(dBm

)

1 202 3 43.2 3.4 3.6 3.8

0 10 20 30 40−90

−85

−80

−75

−70

−65

Time stamp (s)

RSS

I (dB

m)

TheoreticalBF + AKFRawBF

Figure 4: Performance of BF + AKF ltering design. Thezoom-in gure shows a closer view

1 2 3 4 5 6 70

0.2

0.4

0.6

0.8

1

Estimation error (m)

CDF

w./o. ANFw./o. EnvAwarew. ANF + EnvAware

Figure 5: Performance of data preprocessing.

4.3 Performance of EnvAware and ANFWe evaluate the ecacy of EnvAware and ANF for LocBLE by test-

ing their performance separately.

Performance of EnvAware. EnvAware ensures LocBLE to use thecorrect regression model for estimation. To quantify its eect on

overall estimation accuracy, we compare LocBLE’s performance

with the case of removing EnvAware component. In particular, we

tested performance in environments #2-#4 as shown in Table 1,

such as the observer moves from behind the wall (NLOS) to line-

of-sight (LOS) w.r.t. the target; people randomly come in between

during the observer’s movement to form p-LOS paths. We plot-

ted the results in Fig. 5 (a). The removal of EnvAware is found to

increase median error by more than 1m, because LocBLE adapts

itself to environmental changes and updates the regression model

accordingly, thus achieving better accuracy.

Performance of ANF.We used the same data from the EnvAware

experiment, and plotted the performance of removing ANF in Fig. 5.

We observed the accuracy degradation of more than 1.5m due to

the absence of ANF, i.e., ANF plays a critical role in improving

LocBLE’s accuracy. This is because ANF’s smoothing mitigates the

impact of low channel coherence time due to user movements and

environmental changes [9].

5 LOCATION ESTIMATIONLocBLE estimates the target beacon’s location by using a regression-

based data fusion of RSS data and motion sensor data. For location


(𝑥, ℎ) (𝑥 + 𝑏(, ℎ + 𝑑()(𝑥 + 𝑏*, ℎ + 𝑑*)(𝑥 + 𝑏+, ℎ + 𝑑+)

…𝑙. 𝑙( 𝑙* 𝑙+

(0,0) (𝑎(, 𝑐() (𝑎*, 𝑐*) (𝑎+, 𝑐+)

(a) Locating a moving target

(0,0) (𝑎&, 𝑐&)

(𝑥, ℎ)

𝑙& 𝑙+ 𝑙,

(𝑎+, 𝑐+) (𝑎,, 𝑐,)

…𝑙.

BLE beacon

Observer

(b) Locating a stationary target

Figure 6: Two dierent use-cases.

estimation of a stationary target, e.g., nding a lost item with a

BLE beacon, LocBLE performs estimation directly in the observer

device.2For location estimation of a moving target, e.g., locating

a moving smartphone with the BLE beacon function turned on,

LocBLE requires data transmission between the target and the ob-

server. LocBLE gives users options to choose which mode they

need.

Problem formulation.We rst introduce our model from a gen-

eral use-case in which both the observer and the target move

randomly as shown in Fig. 6(a). We assume a coordinate plane

with the origin of the observer’s starting point, and x-axis as the

observer’s starting direction. Our goal is to estimate the target’s

relative location (x,y) in this coordinate. With accurate tracking

of the movement for both the observer and the target, we have

ai , ci as the observer’s real-time x- and y-axis movements at the

same time, the target’s relative x- and y-axis movements as bi , di ,

where i ∈ [0,N ], N is the total number of sample points. Note that

a0 = 0, b0 = 0, c0 = 0, d0 = 0. Then, the corresponding distance

li =√(x + bi − ai )2 + (h + di − ci )2 .

Using the path-loss model [9], we combine movements with RSS

readings based on:

RS = Γ(e ) − 10n(e ) log(li )

l2i = (x + bi − ai )2 + (h + di − ci )

2.(1)

We modied the legacy log-based model to derive the rst equa-

tion of Eq. (1). The key idea is that some parameters may vary with

environment, and hence we use variable e to denote environmental

changes. Here Γ(e ) = P + X (e ), P denotes the power oset that

depends on hardware conguration, X (e ) is the environment noise;

n(e ) is the fading coecient that varies with the environment.

To simplify the notation, we let ϵ = exp

(Γ(e )5n (e )

), η = exp

(−1

5n (e )

)andpi = bi−ai ,qi = di−ci , we can reformulate the second equation

of Eq. (1) to:

p2i + q2

i + 2xpi + 2hqi + x2 + h2 = ϵηRSi . (2)

Note that Eq. (2) shows a similar form to an elliptical regression

problem. So, we form a standard elliptical equation as:

Ap2 + Bq2 +Cp + Dq +G = ρ, (3)

where A = 1

ϵ , B =1

ϵ , C =2xϵ , D = 2h

ϵ , G = x 2+h2

ϵ and ρ = ηRSi .

2In this paper, we dene two types of devices: observer and target.

Our algorithm for deriving the above parameters is based on the

least square regression; specically, we have

P = (XTX )−1XTY , (4)

where P = [1,A,B,C,D,G]′ is the parameter vector, X = [1,p2,pq,q2,p,q] is the data matrix, and Y = [ρ] is the output vector .

In case both the observer and the target are moving randomly, we

assume the observer can communicate with the target. Specically,

after the measurement process, the target will send measurement

data to the observer for processing. If the target remains stationary

(Fig. 6(b)), the problem becomes much simpler and works in a

standalone smart device. Specically, q will become 0.

Solving for the fading coecient. As discussed above, n(e ) can-not be derived explicitly, because the output variable η also contains

n(e ). So, LocBLE determines n(e ) numerically by nding n̂∗ (e ):

n̂∗ (e ) = arдmin

n̂ (e )( L(x̂ , ˆh) − R( n̂(e ), Γ(e ) ) )2, (5)

where L(·, ·, ·) and R(·, ·) are the left- and the right-hand side for-

mula of Eq. (2), respectively.

By solving the corresponding equation, we can estimate the

location for both the target’s stationary and moving cases. Thus,

we can easily infer x̂ andˆh from the parameters derived from the

elliptical regression.

BLEBeacon

A B

C

BLEBeacon

BLEBeacon

Ambiguity 1(𝑥, ℎ%)

Actual location(𝑥, ℎ)

Ambiguity 2(𝑥%, ℎ)

Figure 7: L-shaped movement for locating BLE beacons.

Estimation condence. Let’s revisit the signal propagation model

RS = Γ(e ) − 10n(e ) log(d ). With the actual/estimated coecients

n(e ) and Γ(e ), we can calculate the noise δRS for every RSS sensing

point by subtracting the estimated RSS (R̂S) from the original RSS

(RS), i.e., δRS = RS−R̂S . Ideally, δRS follows a Gaussian distribution

with 0 mean. However, in reality, δRS will not have 0 mean. Assume

δRS ’s mean and standard deviation (std) are µ and σ , respectively.

Mathematically, Gaussian distribution’s σ is robust to the change of

its mean, so we assume σ remains the same and the original Gauss-

ian noise follows N (0,σ ). Therefore, we treat P (µ ) as a probability

(estimation condence), where P (x ) follows N (0,σ ).

5.1 Handling Symmetry AmbiguityDue to the square root process for x andh, our scheme will generate

two possible solutions which are symmetric to the moving path of

the observer. To address this confusion, we design a simple moving


pattern to follow for the observer to rule out ambiguity. LocBLE

focuses on a 2-dimensional space.

L-Shaped Movement Design. Fig. 7 shows an example of 2-D

movement we designed. Suppose the actual location of the target

BLE beacon is at the right-hand side of movement direction

# »AB. The

observer starts to move from point A to point B, using our location

estimation algorithm, one can obtain a result set, {(x ,h), (x ,h′

)},

where (x ,h) is the actual location, and (x ,h′

) is the ambiguous

location at the left-hand side of

# »AB (ambiguity 1 in Fig. 7). To over-

come the ambiguity, the observer continues moving in a dierent

direction, e.g., from B to C . With this movement, our algorithm

can generate another result set, {(x ,h), (x′

,h)}, where (x′

,h) is the

other ambiguity at left-hand side of

# »BC . Thus, to get the location

estimation of the actual beacon, we combine the two estimations

together to disambiguate the false estimations. Specically, we cal-

culate the overlap of two result sets. In case both the observer and

the target move, LocBLE only requires the observer to move with

this pattern.Wewill later discuss how LocBLE estimates short-range

movements.

5.2 Device Motion EstimationLocBLE utilizes motion sensors on smartphones to measure the

moving direction and distance in real-time. To make our motion

tracker independent of phone postures, we use the well-known

coordinate alignment [22] for transforming phone coordinate to

earth coordinate.

To measures the dimension of L-shaped movement, LocBLE

rstlymeasure themoving distance by counting steps recognized ac-

celerometer readings for step count. For measuring turning, LocBLE

uses gyroscope and magnetometer for tracking the degrees of turn.

5.2.1 Measuring moving distance. Measuring the user’s move-

ment with a smartphone has been studied extensively [23–25]. To

determine the moving distance, we rst use the accelerometer to

detect step, and then combine it with step length to get the walking

distance. Our step counter rst smoothes the accelerometer data

by using the moving average lter, then uses a voting algorithm

to detect the peak, which represents the middle status of one gait

cycle. The performance of our step counter is plotted in Fig. 8(a).

To infer the moving steps, we use a similar rationale as in [26].

Specically, we can infer step length by inspecting the step fre-

quency.

5.2.2 Measuring turning angle. To measure turns, we rst an-

alyze gyroscope to identify turning behavior, then use magnetic

heading to infer a specic turning angle. The magnetic eld reading

is known to uctuate in indoor environments, but it is accurate over

a short period time [23]. Specically, to identify turning behavior,

our turn detector inspects gyroscope readings to identify the bump

caused by the turning behavior. Our algorithm can accurately track

the beginning and ending points of a bump. Then, we nd the cor-

responding points in the magnetic heading to get the turning angle,

as shown in Fig. 8(b).

38 39 40 41

−0.4

−0.2

0

0.2

0.4

Time (s)

Am

plitu

de

Raw dataAveraged dataStep detection

(a) Step detection

7 7.5 8 8.5 9 9.5 10−3−2−1

01

Ang

ular

acc

. (ra

d/s2 )

Time (s)

7 7.5 8 8.5 9 9.5 10

50

100

150

Meg

natic

hea

ding

(o )

Time (s)

Raw data

Average

Turn begin

Turn end

(b) Turn detection

Figure 8: Step and turn detection in LocBLE.

Our experimental results show that the accuracy of step-based

moving distance estimation is around 94.77%, and the average angle

estimation error is 3.45◦. To further increase the measurement

accuracy of the L-shaped movement in real-world applications,

LocBLE can avoid the turning angle measurement step by explicitly

asking the user to make a right angle (angle of 90◦) turn when s/he

makes an L-shaped movement.

Algorithm 1 Relative Location Estimation Algorithm

1: INPUT: r so is the observer received RSS, motiono is the observer

motion, similarly (r st ,motiont ) for the target2: Function EstimateLocation (r so ,motiono , r st ,motiont )3: for every time window in r so do4: if The Target is Stationary then5: Detect the observer’s movement (step and direction)

6: else7: Detect both the observer’s and the target’s movements

8: Match the movement to RSS data with timestamp

9: Classify its environment with EnvAware and lter the noise with

ANF

10: if Environment hasn’t change then11: Continue the regression by appending the data

12: else13: Start a new regression with the data

14: Update target location estimation and probability

15: return Target location estimation and its probability

5.3 Summary of Location EstimationAlgo. 1 summarizes our data preprocessing and location estima-

tion. At input, we collect a new data batch every 2–3 seconds with

approximately 20 RSS samples per data batch, and motion data

is also recorded in another data batch. With input data, LocBLE

starts a 3-step estimation process. First, it determines the target’s

status and estimates the observer’s (and the target’s if moving)

step and direction. Second, LocBLE matches the movement with

the RSS data based on the timestamp. Based on the environment

classication, LocBLE turns to the regression by appending the new

data batch if the environment remains unchanged, or to start a new

regression with the data if the environment changes. LocBLE then


0 10 20 30 40 50SignDl ,nWerSRlDWed 3RinWs

−3.0

−2.5

−2.0

−1.5

−1.0

−0.5

0.0

0.5

1.0

1.5

DiIIerenWiDl 3rRcessed 5SS,

SegPenW Win (T1, Size 10)

TDrgeW SeT (T)

0DWched SeT (S1)

1Rise SeT (S2)

0 2 4 6 8Target 6egPent T1

0

2

4

6

8

0atc

hed

6eg

Pen

t 6

11

2StiPaO Path

0 2 4 6 8Target 6egPent T1

0

2

4

6

8

1ois

e 6

eg

Pen

t 6

21 2StiPaO Path

(a) (b) (c) (d)

Figure 9: (a) Sample RSS sequences of 4 beacons; (b) Illustration for segmentation of 3 BLE sequences; (c) DTW cost matrix forsuccessful matching; (d) DTW cost matrix for unsuccessful matching.

applies the adaptive noise ltering. Finally, we return the location

estimation with the corresponding probability by combining these

probabilistic estimations.

6 MULTI-BEACON CALIBRATIONDue to the low cost (usually $6–15 each), the deployment of BLE

beacons can be very exible. Here we will show LocBLE is capable

of rening the localization result with awareness of neighboring

beacons. Specically, LocBLE will rst determine whether or not

other neighboring beacons can form a cluster. If such an opportu-

nity exist, LocBLE will calibrate the estimation result based on RSS

readings from multiple clustered beacons. This method enables the

user to get more accurate result within a single measurement.

6.1 Clustering Correction at TargetNeighboring devices are usually unknown to each other beforehand.

LocBLE aims to group them based only on BLE beacon signals. To

test how RSS signal pattern can be altered by physical layout, we

rstly inspect real-world RSS measurement results of 4 beacons.

Fig. 9(a) shows RSS sequences of onemeasurement, in which beacon

4 is the target device (5m away from observer). Beacons 2 and 3

stay close (0.3m between them) to the target and beacon 1 is not

nearby (4m away). Beacons 2 and 3 are shown to exhibit a similar

pattern of RSS changes. Since our measurement usually requires the

observer to make an “L-shaped” movement, such a pattern becomes

unique in geometry. According to our experimental results, beacons

exhibit disparate RSS trends when they are placed far apart from

each other. Moreover, the target RSS frequency is found to drop

from 8Hz to around 3Hz due to interference. These observations

led us to propose an algorithm based on dynamic time warping

(DTW) [27] to capture the unique trend of RSS readings and cluster

beacons 2, 3, and 4 together. DTW is, in general, used to align and

measure the similarity between two temporal sequences of data. It

formulates the cost matrix based on Euclidean distance between

two datasets and then picks the path with the lowest cost as the

alignment [27].

There are three challenges in using DTW to accurately to match

sequences with the target sequence: 1) dierent devices may have

dierent sensor sampling frequencies andmay sense dierent signal

amplitudes due to chipset and environment variations; 2) DTW has

high time complexity (O (n2)), which is not suitable for large data

sequence matching; 3) possible signicant noise in a data sequence

may lead to incorrect matching for the data sequence. To deal

with these challenges, LocBLE uses a xed window DTW voting

algorithm.

First, to deal with device heterogeneity and environment diver-

sity, our BLE signal processing algorithm lters out high-frequency

noises, and then dierentiates the RSS sequences to avoid using

absolute values.

Second, we reduce the computation overhead by dividing a large

data sequence into small data segments and utilize the lower bound-

ing technique [28]. Fig. 9(b) describes our algorithm for beacons

4, 3, 1. We assume target beacon 4 sequence T and two other se-

quences beacon 3 as S1 and beacon 1 as S2. We rst divide T into

equal-length segments. We found the segment length of 10 to yield

the best tradeo between accuracy and computation complexity.

We then split the other candidate sequences according toTi ’s times-

tamp, and interpolate them to match T ’s segments.

In Fig. 9(b), the data segments are illustrated with red and blue

shaded boxes. Each segment is validated with the lower bounding

technique. Basically, we create a bounding envelope above and

below each target segment using the warping window. Then, we

obtain the squared sum of the distances from every part of the

candidate segment not falling within the bounding envelope, to

the nearest orthogonal edge of the bounding envelope, and check

if it is less than a threshold. The empirical threshold we set for

the segment of 10 data batch is 6.1, which is the same as the DTW

similarity threshold. Only valid data segments can be passed to

the DTW similarity matching. This lower bound testing is 100x

faster than DTW computing for the same size data according to our

experiments. We show the matching for rst segments in Fig. 9(b),

Fig. 9(c) (successful matching) for beacon 3, and Fig. 9(d) (unsuc-

cessful matching) for beacon 1. Our scheme is found to be at least

2x faster than applying DTW directly to the original sequence.

Finally, to obtain the nal matching result, we use straightfor-

ward voting. For every BLE RSS sequence, each segment will be


determined to be successful or unsuccessful based on the above

DTW matching. We then determine a BLE RSS sequence to match

successfully by checking more than a half of the sequence’s seg-

ments to match the target segments.

6.2 CalibrationWe calibrate the target position using the candidate estimations

with probabilistic weights. Note that every target position estima-

tion comes with a probability weight (Sec. 4), and our calibration

works with this estimation. Suppose there are N candidate target

positions with probability pi for each candidate i , then we use the

normalized weightpi(∑Nj=1 pj

) and compute the weighted sum of

the candidate positions as the nal estimation. The detail of our

algorithm can be found in Algo. 2.

Algorithm 2 ClusteringCalibration Algorithm

1: INPUT: Pre-processed LocBLE’s RSS sensing sequences for dierent

devices S1, . . . , SN and observer–target sequence T , mo is the ob-

server’s motion andmt is the target’s motion

2: Function ClusteringCalbration (S1, . . . , SN , T ,mo ,mt )

3: Init Tarдet_Posit ions as an empty list

4: Split T into segments T1, . . . , Tm with 10 points each

5: for every sequence Si do6: Split Si into segments Sjk (k in (1,m)) according to Ti7: Initmatched_count = 0

8: for every pair Sik and Tk (k in (1,m)) do9: if Sik passes lower bound threshold and Sik satises DTW simi-

larity threshold then10: matched_count ++11: if matched_count > m

2then

12: Set tarдet_pos = EstimatePosit ion (r si , r s,mo,mt )13: Add tarдet_pos to Tarдet_Posit ions14: Compute the weightwi for each tarдet_posi in Tarдet_Posit ions

15: return Weighted Sum of Tarдet_Posit ions

7 EVALUATIONThis section provides the details of our implementation and experi-

ment setting, a demonstration of using LocBLE in action, evaluation

of LocBLE’s performance and overhead in various environments.

7.1 ImplementationWe have implemented LocBLE on both Android and iOS devices.

LocBLE runs on recent generations of iOS (version iOS 10.2.1) and

Android (Android 6.0) devices, including iPhone 5s/6/6s/7, iPad Air,

Nexus 6P. For iOS, LocBLE relies on CoreBluetooth framework to

obtain the BLE RSS data and CoreMotion framework to get device

sensing information. Since iOS 5.0, Apple no longer allows an iOS

device to read unknown Bluetooth device’s UUID, LocBLE requires

the user to add the target device UUID to the app. It uses library

SWIX [29] for the regression and machine learning classier. For

Android, we use the Bluetooth package for BLE scan and sensor

package for motion sensing data. For the moving target scenario,

we use UPnP [30] protocol to enable direct communication between

two devices.

As shown in Fig. 10(a), LocBLE’s GUI contains both measureand navigation modes. In measure mode, the user selects the target

device and then follows the LocBLE’s instruction arrow for an L-

shapemove. The instruction arrow grows based on the step detector,

while LocBLE estimates the user movement with sensor data. Upon

completion, LocBLE computes and shows the relative position of

the target device. In navigationmode, LocBLE provides instructions

based on the measured target position so that the user can nd the

target device. In LocBLE’s current implementation, navigation is

based on standard dead-reckoning with a step counter [31].

7.2 Methodology and MetricsExperimental Setup. Our evaluation focuses on iBeacon devices,

including various models of o-the-shelf smart devices and dedi-

cated BLE beacons, including iPhone 5/5s/6/6s, Estimote beacons [32],

and RadBeacon USB dongle [33]. We congured the BLE beacons

to broadcast at 10Hz in order to be consistent with sampling rate.

In our experiments, we assumed the static targets except for the

moving target experiment. Our experiment setting includes vari-

ous real-life environments. As shown in Table 1, our experimental

setting includes 9 dierent environments (indexed from 1 to 9) — 8

indoor and 1 outdoor environments. The settings also include line-

of-sight (LOS) and non-line-of-sight (NLOS) scenarios in which

direct paths are blocked by furniture, store/shop racks, and hu-

man bodies. To test the practicability of LocBLE, our indoor test

environment did not rule out WiFi access points.

Metrics and Ground Truth. The metric of our interest is the tar-

get location estimation error, or the dierence in distance between

the target’s estimated location and the ground truth. In the case

of a moving target, we measured the target location estimation

error at its initial location. In order to accurately compare LocBLE’s

optimization algorithms against other alternatives, we performed a

trace analysis. For this analysis, we collected the traces in 9 dier-

ent environments (Table 1) covering most of our daily use-cases.

Each trace had distances ranging from 3 to 16m and around 40

BLE data samples. Our dataset has nearly 300MB over the distance

measurements of 1,000m. This dataset is used to evaluate LocBLE

as described below.

7.3 Demonstrate LocBLE in ActionDemo [34] shows an example process of LocBLE navigation func-

tionality. We randomly place the target Estimote beacon in an oce

and then use LocBLE to measure the location and navigate to nd

the target device. We measured the distance from LocBLE’s nav-

igation destination to the actual beacon location as the overall

error. We conducted 20 runs with the distance to the target device

ranges from 4 to 12m and show the overall error CDF in Fig. 10(b).

The median overall error is 1.5m, 75 percentile error is 2m and

the maximum error is less than 3m. This experiment demonstrates

the feasibility of measuring the target BLE device’s location and


Index 1 2 3 4 5 6 7 8 9

Image

Name

Meeting

room

Hallway Bedroom

Living

room

Restaurant Store Labs Hall

Parking

lot

Scale 5×5 (m2) 8×3 (m2

) 7×7 (m2) 7×7 (m2

) 9×10 (m2) 9×10 (m2

) 8×10 (m2) 9×11 (m2

) 16×15 (m2)

Acc. (m) 0.8 ± 0.2 1.4 ± 0.3 1.4 ± 0.4 1.6 ± 0.3 1.6 ± 0.4 1.8 ± 0.6 2.3 ± 0.5 2.1 ± 0.5 1.2 ± 0.5

Table 1: Details of experimental environments.

(a) (b)

Figure 10: (a) LocBLE App UI; (b) overall error in action

navigating to the target with the measured location, and handles

various practical scenarios with reasonably good accuracy.

7.4 Performance Analysis of LocBLE inDierent Environments

We now evaluate LocBLE’s performance for both stationary and

moving target devices in various environments.

7.4.1 Stationary Target. A main application of LocBLE is to lo-

cate stationary BLE beacons. To guarantee the generality of our

experiment, we tested LocBLE’s performance in environments #1 –

#6 as shown in Table 1. The original distance between the target

and the observer in dierent environments are 4.5m, 6.4m, 6.7m,

6.8m, 9.1m and 7.9m, respectively. We plotted the mean error for x ,

h and absolute distance

√(x2 + h2) in Fig. 11(a). LocBLE is found

to have less than an absolute distance error of 1m for the meeting

room. In more challenging environments, LocBLE is found to have

less than 2.4m absolute distance error due to the multi-path eects.

LocBLE is found capable of providing the target’s actual location

estimation, i.e., x and h, while no existing solution has this capabil-

ity. Instead, the existing solutions focus on range estimation with

BLE proximity capability. So, we choose the best ranging app called

Dartle [35] for comparison. We measured it under the same settings

and plot its performance in Fig. 11(a) and nd LocBLE achieving

30% less error than Dartle app’s ranging estimation.

0

1

2

3

4

5

6

1 2 3 4 5 6Es

timat

ion e

rror

(m)

x est.h est.LocBLE abs.Dartle app

(a) Stationary

0 2 40

0.2

0.4

0.6

0.8

1

CDF


Test 1

Test 2

(b) Moving

Figure 11: Performance of stationary andmoving device sce-narios. In (a), the x-axis shows the environment index

Performance analysis in dierent environments. Table 1 sum-

marizes the LocBLE’s performance from all ranges in locating sta-

tionary beacons. The results in the 5th

row show LocBLE’s mean

accuracy with a 75%-condence interval. There are two main take-

aways from this study of LocBLE’s performance. First, LocBLE

shows better performance when a LOS path exists. The meeting

room environment shows the best performance with <1m accuracy,

while the store and lab environment show the worst performance

due to the existence of highly reective blocking objects (e.g., high

market and server racks) in the signal propagation path. Second

LocBLE shows stable performance with dierent environment set-

tings where NLOS paths exist. Specically, environments #2 – #6

show similar accuracies. As we will show in Sec. 4.3, our EnvAware

helps LocBLE rene the estimation results when the environment

changes.

7.4.2 Moving Target. Another feature of LocBLE is the abilityto estimate the relative location of a moving target. To the best of

our knowledge, there doesn’t exist any solution for this purpose. In

order to test LocBLE’s performance for a moving target, we set up

experiments with two users, one as the observer and the other as

the target. Each user holds an iOS device with LocBLE activated. We

xed the start point for both users and pre-dene several moving

directions for each user. The two users moved in dierent directions


0

1

2

3

4

2.8 5.6 8.4 11.2 14 > 15

Estim

atio

ner

ror(

m)

Absolute distance (m)

> 3.5 m

(a) Dierent target distance

0123456

17 14 11 9 6 3

Estimationerror(m)

Absolute distance (m)

(b) Navigation performance

Figure 12: LocBLE’s performance for dierent target dis-tances and navigation

and at dierent speeds. Moreover, both users move during the mea-

surement. After measurement, the target transfers RSS and motion

data traces to the observer for distance estimation. We tested the

performance in environments #9 (test 1) and #8 (test 2), more than

40 experiments have been conducted in each environment. Test 1

changed distance from 3m to 9m, whereas in test 2 did from 3m

to14m. We plot the CDF of error in Fig. 11(b). The results indicate

the accuracy of less than 2.5m for more than 50% of data we col-

lected. There are two sources of error aecting the performance.

First, the BLE signal blockage changes too fast for the environment

classier to capture during the human’s walking. Second, the er-

ror in movement estimation accumulates, especially in estimating

the moving direction of two users. Thus, LocBLE can expect more

accurate estimation with a better environment adapter and ne

movement tracking.

7.5 Performance Analysis with ChangingTarget Distance

Target Distance on Measurement Accuracy. Due to the log-

based degradation of RF signal, a natural question is: how does

LocBLE’s performance change with dierence target distances? As

discussed in literature [17], BLE-based sensing shows valid prox-

imity estimation within an about 15m range. To ensure the same

environment for a fair comparison, we conducted experiments in

a suciently large outdoor parking lot. We collected the estima-

tion results with LocBLE placed at 11 dierent testing points, two

adjacent points of which were separated by 2.8m. For every point,

we repeated the experiment 5 times and plotted the mean error

in Fig. 12(a). LocBLE is found able to achieve around 1m accuracy

within 5.6m and <3m accuracy within an 11.2m range. However, if

the distance is over 14m, the performance degrades signicantly

to more than 3m which is consistent with our expectation. This

is because a log-based propagation model has a minor decrease at

long distances.

Target Distance on Navigation. LocBLE provides a navigation

service for the observer to nd the target. So, we assess LocBLE’s

navigation performance with a use-case. Basically, an observer who

is 16.5m away from the target rst estimates the target’s location

0 2 40

0.2

0.4

0.6

0.8

1


CDF

9 Hz8 Hz6.5 Hz5.5 Hz

(a) Changing sampling frequency

1 2 3 40

0.2

0.4

0.6

0.8

1


CDF

100%80%70%50%

(b) Varying data length

Figure 13: Comparison of estimation results for dierentsampling frequencies and data quantities.

with LocBLE. Then, the observer follows the guidance of LocBLE

to approach the target. We recorded the estimation accuracy at

dierent locations. We repeated the experiments 3 times and show

the mean error at dierent points in Fig. 12 (b). In the beginning,

due to the long distance and insucient BLE sensing data, the error

reaches nearly 5m. LocBLE’s performance improves as the observer

approaches the target, especially when the distance reduces to 3m,

the error drops to 1m.

7.6 LocBLE’s Performance with DierentSettings

In real-life applications, it is unlikely to keep settings ideal and

homogeneous due to dierent hardware congurations and user

behavior. Thus, it is necessary to study whether LocBLE will vary

drastically with settings. Specically, we focus on evaluating how

sampling frequency, user’s walking distance, and types of BLE

beacon will impact on LocBLE’s performance.

7.6.1 What sampling frequency is suicient? Devices with dier-

ent BLE chipsets and OSes may sense or broadcast BLE at dierent

frequencies. For example, the sampling rate is 9Hz for iPhone 6s

and 8Hz for Nexus 6P. We quantify the impact of sampling fre-

quency on the estimation accuracy by re-sampling our data at a

lower frequency. Our original data sampling rate is about 9Hz with

iOS devices and the average length for every measurement are

about 40 points. We conducted experiments in environment #2–#4,

and compared frequencies 8Hz, 6.5Hz, and 5.5Hz (by inserting an

idle delay between two consecutive scans). The result shown in

Fig. 13(a) indicates that with a lower sampling frequency of data

points being used for LocBLE, the medians of estimation results

remain stable, but in the worst case, the lower sampling rate may

degrade the performance, because a lower sampling frequency still

ensures data completeness, but more susceptible to noises.

7.6.2 How far does the observer need to walk? Ideally, given an

accuracy requirement, a shorter walk will yield better user experi-

ence. In LocBLE, we require the observer to walk 3.5–5m in total

for an L-Shape path. So, how does the measurement of walking

distance aect the overall estimation accuracy? To answer this


0

0.5

1

1.5

2

iOS Rad Beacon Estimote

Estim

atio

ner

ror(

m)

Figure 14: Performance on dierent beacons

question, we evaluated its impact by varying the length of the mea-

surement data. The CDF of error is plotted in Fig. 13 (b). We nd the

performance remaining stable after reducing the data to 80% of the

original data, but starts to degrade at 70%, and becomes much worse

at 50%. This indicates that LocBLE needs at least 80% of data, that

is ∼3m walking distance, to capture the signal characteristics and

provide accurate estimation. In practice, LocBLE asks the observer

to take around 4–6 steps in the measurement, usually taking about

3–5s.

7.6.3 Does BLE beacon type maer? To verify this, we used

three common types of BLE beacons in the market as the targets

and measured the estimation errors in environment #2. Fig. 14

shows the average errors for using three dierent types of BLE

beacons as the target. Dedicated BLE beacons (RadBeacon [33] and

Estimote [32]) have slight advantages over smart devices integrated

beacons, as the chips in smart devices are built more compactly.

Regardless of this minor dierence, the experimental results show

that LocBLE doesn’t depend on specic BLE devices.

7.7 Clustering CalibrationWe evaluate the ecacy of our clustering scheme on enhancing

accuracy. We conducted experiments in scenarios 7 and 8 as shown

in Table 1 including a lab environment with a concrete wall block in

the transmission path, and a hall environment with a construction

in between. Due to heavy blockage and long distance, the accu-

racy for one target with a single beacon averages only 3m. As we

increase the number of beacons, the accuracy improves as shown

in Fig. 15. With 6 beacons, LocBLE reduces the error by half. A

single device’s estimation has a low estimation condence and in-

dicates low accuracy in these environments. That is, with LocBLE’s

clustering calibration, the estimations from those neighboring de-

vices compensate the noise in the challenging environments, and

improve the overall accuracy.

7.8 System OverheadLocBLE doesn’t require any hardware or OS modication and has

low energy consumption at both the target and observer sides. To

quantify this, we logged the energy consumption as a time series by

instrumenting LocBLE on XCode [36]. The result shows that LocBLE

increases CPU usage by 14% and energy consumption by 12%, which

is slightly higher than the existing ranging app (Dartle [35]) which

0

1

2

3

4

1 2 4 6Estim

atio

n er

ror (

m)

Number of beacons

Lab Hall

Figure 15: Calibration performance in two environments

has 11.3% and 11% increases, respectively. We also expect similar

energy consumption on Android.

8 RELATEDWORK8.1 Nearable TechnologiesThe BLE proximity beacon technology has received signicant at-

tention for its potential for enriching human–environment interac-

tions. For example, manufacturers, such as Apple [37], Google [38]

and open-sourced community [39], proposed their own BLE proxim-

ity beacon specications. However, existing pervasively-deployed

BLE beacons are only able to provide coarse-grained proximity.

LocBLE is a fundamental enhancement of existing nearable tech-

nologies by enabling estimation of the target beacon’s location

with a single smartphone, without modication to the smartphone

and the proximity beacon. These features extend the usability of

existing nearable beacons.

8.2 Transmitter Tracking with RF SignalsTransmitter localization based on RF signals has been an active

area of research, achieving high indoor localization accuracy. In

general, to locate the RF signal transmitter’s location, existing re-

search eorts focus on angle-of-arrival (AoA) [40, 41], time-of-ight

(ToF) [5, 42], or RSS [43, 44] analysis. To measure AoA, the RF re-

ceiver needs to have an antenna array for analyzing phase dier-

ences. To measure ToF, the RF receiver should have a nanosecond-

level sampling rate to achieve ne-granularity, which is infeasible

on mobile devices. Thus, these techniques are infeasible for low-

power BLE beacons.

RSS-based approaches, e.g., Borealis [45], is designed for locatinga WiFi AP in outdoor environments without dedicated hardware.

However, [45] requires OS modication on smartphones to support

WiFi RSS scan on a single channel. Unlike AoA, ToF and RSS based

device tracking, FindMiMo [46] is a tracking system for nding

a mobile phone by utilizing various logging information, such as

AP identier, GPS, and motion sensors. This system is infeasible

to locate low-power BLE beacons due to the limited information

the user can get from BLE beacon advertisements. In contrast with

existing approaches, LocBLE does not require OS modication on

smartphones or BLE beacons.


802.11 mc [47] enabled ne-grained ToF measurement at receiver

side. However, this protocol is designed for WiFi devices, thus not

feasible in IoT use-cases.

8.3 Environment EstimationPath-loss exponent (PLE) estimation has been studied extensively in

the area of wireless sensor networks. Unlike the various estimation

schemes in literature [10, 11], LocBLE employs a novel path loss

exponent estimation based on data fusion of RSS readings and

motion sensors. It enables light-weight path loss estimation on

mobile devices.

9 DISCUSSION AND FUTUREWORKWe discuss LocBLE’s usability and it’s limitations in real-life scenar-

ios. Finally, we showcase LocBLE’s extensibility in future works.

9.1 On Accuracy of LocBLEThe design of LocBLE strives to achieve good accuracy by miti-

gating RSS signal uctuations. The accuracy is around 1.8m in an

indoor area, which is less accurate than other dedicated localiza-

tion systems (e.g., WiFi AP with a big antenna array). However,

the exibility (using commodity BLE beacon) and light-weighted

(requiring only smartphone to operate) features together make

LocBLE a unique and eective choice in the IoT paradigm (e.g., helpusers nd a lost item faster).

9.2 LimitationsLast meter navigation. LocBLE can navigate user to the target

within 2m even from more than 10m away. However, it is not easy

for LocBLE to make further improvement due to the nature of RSS.

From our experiments, we observed that the Bluetooth proximity

actually demonstrates fairly good accuracy within 2m. Therefore,

if we incorporate proximity in LocBLE, we will be able to bring

accuracy under 1m or even cm level. We leave this as our future

work.

L-shaped measurement. LocBLE requires the observer to move

“L” shape in the measurement stage. Although LocBLE only requires

several steps of walk, it may still be inconvenient in some space

limited area. To solve this diculty, the observer may just walk

straight and leave the symmetry problem to the navigation stage.

During the last turn in navigation, we will know whether the ob-

server is in a correct direction and correct him accordingly. Such

design and implementation will be our future work.

Evaluation in crowded environments. Although LocBLE has

shown to yield promising result given interference fromWiFi in the

vicinity, more experiments should be conducted in an environment

with saturated interferences. For example, in a shopping mall where

pedestrians’ BLE signals and the surrounding BLE beacons create

interferences and aect RSS readings. This scenario may require

LocBLE to tune location estimation based on channel interference.

9.3 Extensibility of LocBLE3-dimensional measurement. The current LocBLE is designed

to show beacons’ locations in a 2-D space. However, in real-world

scenarios, users may prefer a 3-D location information for a more

engaging experience (e.g., locating tagged devices more accurately

on the user’s AR device). 3-D localization can be done by modifying

our data fusion and L-shaped movement. We leave the detailed

design and evaluation of this as our future work.

Compatibility with Bluetooth 5.0. Bluetooth 5.0, as the next

standard of BLE technology, oers four major improvements, in-

cluding wider coverage [48]. This feature will enhance LocBLE’s

performance while keeping it still compatible with the upcoming

Bluetooth 5.0.

10 CONCLUSIONWe have presented LocBLE, a novel framework for locating BLE

beacons on smartphones. It is shown to be able to achieve m-level

accuracy with only commodity smartphone and proximity BLE bea-

cons. By recognizing the environmental changes and multi-nodes

scenarios, LocBLE can signicantly enhance its location estimation

accuracy. It provides a practical location estimation method for

low-power IoT devices, enabling many IoT apps that have not been

possible before.

REFERENCES[1] How beacons will inuence billions in us retail sales. http://www.businessinsider.

com/beacons-impact-billions-in-reail-sales-2015-2.

[2] Health media network announces partnership with gimbal, inc.

https://www.gimbal.com/press-releases/health-media-network-announces-

partnership-with-gimbal-inc-creating-the-rst-and-largest-healthcare-

proximity-beacon-network-in-the-u-s/.

[3] Waze launches bluetooth beacons to avoid tunnel blackouts.

https://techcrunch.com/2016/09/21/waze-launches-bluetooth-beacons-to-

avoid-tunnel-blackouts/.

[4] How beacons can reshape retail marketing.

https://www.thinkwithgoogle.com/articles/retail-marketing-beacon-

technology.html.

[5] Souvik Sen, Jeongkeun Lee, Kyu-HanKim, and Paul Congdon. AvoidingMultipath

to Revive Inbuilding WiFi Localization. In ACM MobiSys, 2013.[6] Ramsey Faragher and Robert Harle. An analysis of the accuracy of bluetooth low

energy for indoor positioning applications. In Proceedings of the 27th InternationalTechnical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+2014), Tampa, FL, USA, volume 812, page 2, 2014.

[7] W. W. L. Li, R. A. Iltis, and M. Z. Win. A smartphone localization algorithm using

rssi and inertial sensor measurement fusion. In IEEE Global CommunicationsConference (GLOBECOM), pages 3335–3340, Dec 2013.

[8] Gaddi Blumrosen, Bracha Hod, Tal Anker, Danny Dolev, and Boris Rubinsky.

Enhanced calibration technique for rssi-based ranging in body area networks.

Ad Hoc Netw., 11(1):555–569, January 2013.

[9] David Tse and Pramod Viswanath. Fundamentals of Wireless Communication.Cambridge university press, 2005.

[10] Guoqiang Mao, Brian D.O. Anderson, and BarÄśÅ§ Fidan. Path loss exponent esti-

mation for wireless sensor network localization. Computer Networks, 51(10):2467– 2483, 2007.

[11] J. Koo and H. Cha. Localizing wi access points using signal strength. IEEECommunications Letters, 15(2):187–189, February 2011.

[12] Jue Wang and Dina Katabi. Dude, where’s my card?: Rd positioning that works

with multipath and non-line of sight. In ACM SIGCOMM, 2013.

[13] Longfei Shangguan, Zheng Yang, Alex X. Liu, Zimu Zhou, and Yunhao Liu.

Relative localization of rd tags using spatial-temporal phase proling. InUSENIXNSDI, 2015.

[14] NFCWhite Paper. https://www.ecma-international.org/activities/Communications/tc32-

tg19-2005-012.pdf.

[15] Title 47 cfr part 15. http://www.ecfr.gov/cgi-bin/47cfrv1_02.tpl.

http://www.businessinsider.com/beacons-impact-billions-in-reail-sales-2015-2

http://www.businessinsider.com/beacons-impact-billions-in-reail-sales-2015-2

http://www.ecfr.gov/cgi-bin/47cfrv1_02.tpl


[16] Bluetooth 5.0 core specication, transmitter characteristics, page 2536.

https://www.bluetooth.com/specications/adopted-specications.

[17] BLE Specication. https://www.bluetooth.com/specications.

[18] Bluetooth low energy beacons. http://www.ti.com/lit/an/swra475a/swra475a.pdf.

[19] Bcm4339 datasheet. http://www.cypress.com/le/298016/.

[20] SVM in sklearn module. http://scikit-learn.org/stable/modules/svm.html.

[21] Anonymity. Locating and Tracking BLE Beacons with Smartphones. Technical

report, Please refer to PC Chair for a Technical Report copy.

[22] Dongyao Chen, Kyong-Tak Cho, Sihui Han, Zhizhuo Jin, and Kang G. Shin.

Invisible sensing of vehicle steering with smartphones. In ACM MobiSys, 2015.[23] Alex T. Mariakakis, Souvik Sen, Jeongkeun Lee, and Kyu-Han Kim. Sail: Single

access point-based indoor localization. In ACM MobiSys, 2014.[24] Nirupam Roy, He Wang, and Romit Roy Choudhury. I am a smartphone and i

can tell my user’s walking direction. In ACM MobiSys, 2014.[25] Yuanchao Shu, Kang G Shin, Tian He, and Jiming Chen. Last-mile navigation

using smartphones. In ACM MobiCom, 2015.

[26] Fan Li, Chunshui Zhao, Guanzhong Ding, Jian Gong, Chenxing Liu, and Feng

Zhao. A reliable and accurate indoor localization method using phone inertial

sensors. In ACM UbiComp, 2012.[27] Donald J. Bemdt and James Cliord. Using dynamic time warping to nd patterns

in time series, 1994.

[28] Chotirat Ann Ratanamahatana and Eamonn Keogh. Everything you know about

dynamic time warping is wrong. In Third Workshop on Mining Temporal andSequential Data, 2004.

[29] swix. http://stsievert.com/swix/.

[30] Mohamed Boucadair, Reinaldo Penno, and Dan Wing. Universal plug and play

(upnp) internet gateway device-port control protocol interworking function

(igd-pcp iwf). 2013.

[31] Azkario Rizky Pratama, Risanuri Hidayat, et al. Smartphone-based pedestrian

dead reckoning as an indoor positioning system. In System Engineering and

Technology (ICSET), 2012 International Conference on, pages 1–6. IEEE, 2012.[32] Estimote Beacons. http://estimote.com/.

[33] Radbeacon usb dongle. http://www.radiusnetworks.com/.

[34] LocBLE Demo. https://goo.gl/6MJOeX.

[35] Dartle iBeacon Locator. https://cloud.dartle.io/.

[36] XCode IDE for iOS development. https://developer.apple.com/xcode/ide/.

[37] Apple iBeacon. https://developer.apple.com/ibeacon/, 2013.

[38] Eddystone Beacon. https://developers.google.com/beacons/.

[39] Altbeacon. http://altbeacon.org/, 2015.

[40] Jie Xiong and Kyle Jamieson. ArrayTrack: A Fine-grained Indoor Location System.

In USENIX NSDI, 2013.[41] Swarun Kumar, Ezzeldin Hamed, Dina Katabi, and Li Erran Li. LTE Radio Ana-

lytics Made Easy and Accessible. In ACM SIGCOMM, 2014.

[42] Manikanta Kotaru, Kiran Joshi, Dinesh Bharadia, and Sachin Katti. SpotFi:

Decimeter Level Localization Using WiFi. In ACM SIGCOMM, 2015.

[43] P. Bahl and V. N. Padmanabhan. Radar: an in-building rf-based user location and

tracking system. In IEEE INFOCOM, 2000.

[44] Moustafa Youssef and Ashok Agrawala. The horus wlan location determination

system. In ACM MobiSys, 2005.[45] Zengbin Zhang, Xia Zhou, Weile Zhang, Yuanyang Zhang, Gang Wang, Ben Y.

Zhao, and Haitao Zheng. I am the antenna: Accurate outdoor ap location using

smartphones. In ACM MobiCom, 2011.

[46] Hyojeong Shin, Yohan Chon, Kwanghyo Park, and Hojung Cha. Findingmimo:

Tracing a missing mobile phone using daily observations. In ACM MobiSys, 2011.[47] E. Au. The latest progress on ieee 802.11mc and ieee 802.11ai [standards]. IEEE

Vehicular Technology Magazine, 11(3):19–21, Sept 2016.[48] Bluetooth 5.0 standard. https://www.bluetooth.com/specications/adopted-

specications.

https://www.bluetooth.com/specifications

http://stsievert.com/swix/

http://estimote.com/

https://goo.gl/6MJOeX

https://cloud.dartle.io/

https://developer.apple.com/ibeacon/

https://developers.google.com/beacons/

http://altbeacon.org/

locating and tracking ble beacons with smartphones

Documents